Get answers to some of the most frequently asked questions about ECHELON software and our reservoir simulation tools.
How large a model can you fit on a GPU?
For black oil models memory usage is about 3GB per million cells which yields upper limits around 10 million cells for the 32GB NVIDIA Volta. Typical compositional models with 6 to 8 components can range up to 4 million cells on the NVIDIA Volta. With 18 to 20 components the upper limit on the model size is about 1.5 million cells.
Which preconditioners and solvers are being used by ECHELON?
ECHELON reservoir simulation software employs a GMRES solver in the outer loop with a constrained pressure residual (CPR) preconditioner. Algebraic multigrid (AMG) is used for the first linear stage, while a variety of second-stage preconditioners are available and are automatically selected based on model properties.
Can you explain how the data gets to the GPU if not through the CPU bottleneck?
Moving all the data to describe even a very large model into GPU memory takes only a few seconds. A key point however is that ECHELON reservoir simulation software retains model data resident in GPU memory for the duration of the simulation, so this data movement penalty is paid only once, and thus does not constitute a bottleneck. In contrast, hybrid approaches that accelerate only a portion of the simulation (e.g. the linear solver) with GPUs must move data between CPU and GPU every solver iteration during the simulation, which significantly limits the speed benefits that the GPUs can provide.
What special features in ECHELON software are available for modeling unconventional reservoirs?
For representing hydraulic fractures, users of ECHELON reservoir simulation software have used 1) local grids using our easy to use LGR template, 2) very fine grids with a large number of cells and enhanced permeability for fractures or 3) dual-permeability capabilities. Note that ECHELON software supports compaction tables as is required for approximating fracture closure.
Does ECHELON software have capabilities for user defined control of wells and well groups?
Yes. Our current ECHELON 2.0 software release has support for user defined controls for wells and well groups.
What are the hardware and software requirements for running ECHELON software?
ECHELON software will run with any NVIDIA GPU made after 2013 on machines running Linux or Windows 10. Commercial releases support x86 CPUs, but binaries for IBM POWER and ARM can be made available upon request.
Can you run ECHELON software on multiple GPUs?
ECHELON software was designed from the beginning to handle the largest of reservoir models and can be parallelized across multiple GPUs resident in the same or multiple nodes.
What shortcuts are taken to improve performance in ECHELON software relative to other reservoir simulators?
None. ECHELON reservoir simulation software solves the same governing equations using the same fidelity and convergence criteria as industry-standard legacy simulators to generate equivalent results in less time by combining state-of-the-art algorithms with the extreme performance of GPUs.
Are expensive high-end GPUs required to achieve great performance on ECHELON software?
Although the higher-end GPUs give the best performance, there are many low cost GPUs that will give excellent performance with ECHELON reservoir simulation software. The key aspect to consider is the bandwidth on the GPU, which is directly proportional to the performance achieved by ECHELON.
Are special solvers used to take advantage of GPU architecture?
No special solvers are used, however ECHELON software implements state-of-the-art preconditioning methods, including CPR-AMG, in novel ways to take maximum advantage of the GPU architecture.
A lot of reservoir simulations are still conducted on workstations rather than clusters. Does the GPU still have an edge over the CPU in those setups?
Yes. There are many low cost (under $1k) desktop GPUs available that offer exceptional performance with ECHELON software. This gives ECHELON an enormous edge over CPU simulators by providing its extreme performance without the need for massive upgrades in hardware.
What does GPU stand for?
GPU stands for 'graphics processing unit'.
Are GPUs useful for compositional modelling?
Yes. A common misconception is that GPUs are not effective at handling complex reservoir simulation problems such as compositional. ECHELON compositional gives the same exceptional performance as you see in its black oil simulations.
Does ECHELON software work with third-party geomodeling software, such as Petrel™, RMS™, etc.?
ECHELON reservoir simulation software can work with any software package that exports into Eclipse™ format.
What tools can be used to build models and visualize reservoir simulation results for ECHELON software?
ECHELON software is Eclipse™-compatible, so any of the many pre and post simulation tools can be used to build models and visualize reservoir simulation results for ECHELON.